Homodimerization of amyloid precursor protein and its implication in the amyloidogenic pathway of Alzheimer's disease.

We reported previously that the carbohydrate domain of the amyloid precursor protein is involved in amyloid precursor protein (APP)-APP interactions. Functional in vitro studies suggested that this interaction occurs through the collagen binding site of APP. The physiological significance remained unknown, because it is not understood whether and how APP dimerization occurs in vivo. Here we report that cellular APP exists as homodimers matching best with a two-site model. Consistent with our published crystallographic data, we show that a deletion of the entire sequence after the kunitz protease inhibitor domain did not abolish APP homodimerization, suggesting that two domains are critically involved but that neither is essential for homodimerization. Finally, we generated stabilized dimers by expressing mutant APP with a single cysteine in the ectodomain juxtamembrane region. Mutation of Lys(624) to cysteine produced approximately 6-8-fold more A beta than cells expressing normal APP. Our results suggest that amyloid A beta production can in principle be positively regulated by dimerization in vivo. We suggest that dimerization could be a physiologically important mechanism for regulating the proposed signal activity of APP.

APP is a ubiquitous glycoprotein expressed highly in neurons (4). APP-deficient mice are viable and develop normally, but they display minor defects that differ according to the null strain (5); e.g. copper levels are increased in the cerebral cortex and liver of APP and APLP2 knockout mice (6). Thus, APP is not required for the expression of a critical cell function but may be involved in the modulation of neuronal functions at the cellular level. The structure of APP resembles a cell surface receptor (7). APP has a function in cell-cell and cell-matrix interactions that is evident by binding to heparin and collagen (8 -11). Soluble forms of APP promote neurite outgrowth and cell adhesion and are neuroprotective (12). Surprisingly, the ectodomain of APP is phosphorylated within the acidic region in the N-terminal half of the molecule (13,14). The phosphorylation sites were mapped to serine residues 189 and 206 within the zinc(II) binding site of APP (15)(16)(17). The ectodomain phosphorylation of APP is inhibited by heparin, most likely by steric hindrance through binding of heparin to the N-terminal domain (11,17). Likewise, heparin also inhibits APP-APP interaction in vitro (8).
It has been shown that in brain a distinct percentage of APP is present on the cell surface as a membrane protein of type I. This cell surface APP may mediate the transduction of extracellular signals into the cell via its C-terminal tail (18,19). Involvement of APP in neuronal development, synaptogenesis, and synaptic plasticity indicated that the observed function is not restricted to secreted APP, raising the possibility that some aspects of synaptic plasticity are mediated by cell-associated APP (1,20).
APP is found on the cell surface in neurons (19,21) possessing a neurite outgrowth activity (22). It was shown that APP binds to the brain-specific signal-transducing G protein G 0 (23). Interestingly, APP mutants of familial Alzheimer's disease can cause G 0 -mediated apoptosis in neuronal cells, and these same familial Alzheimer's disease mutants cause the intracellular accumulation of C100 (1,24). Dimerization of receptors induced by ligand binding is a common means of triggering signal transduction in mammalian cells (25). Dimerization appears to play a role in cytokine receptor signaling (26), and conformational changes brought about by receptor oligomerization in response to ligand binding are likely to activate the tyrosine kinase receptors (25). Evidence that dimerization of proteins involved in neurodegenerative processes could play a role in the disease state is derived from current models of the role of PrP in scrapie pathogenesis (27). Oligomeric forms of PrP are believed to facilitate a more rapid conversion of PrP c into PrP Sc and were observed in scrapie-infected hamster brains (28).
This encouraged us to investigate whether cellular APP exists as a homodimer and whether dimerization can promote or attenuate the A␤ production in the ␤-secretory pathway of APP. Based on surface cross-linking studies, we found that APP homodimerizes efficiently in the ER and Golgi and is secreted as a soluble protein from eukaryotic cells and from yeast cells as a recombinant protein. A mutationally induced dimerization of cellular APP produced a 7-fold increase of soluble A␤, which is central to the pathogenesis of the disease. This suggests that dimerization-mediated regulation of APP processing is likely to be physiologically relevant. APP may exist as a homodimer on the cell surface. The homodimer is stabilized as an inactive dimer by multiple dimerization interfaces. A dimerization in this manner would provide a potential mechanism for a negative regulation of proposed APP functions and a concomitant increase of amyloid formation.

Expression Vectors and Site-directed Mutagenesis for Receptor
Dimerization Mutants-APP mutations were introduced by polymerase chain reaction amplification using the megaprimer method (29). Two outside primers spanned the BglII (forward, TCG GCC TCG TCA CGT GTT C) and XmnI sites (reverse, CAA CTG GCT AAG GGG CTA TGT G) in APP while the reverse mutagenic primers introduced mutations at position 624 of APP 695 (K624A, CCA ATG ATT GCA CCT GCG TTT GAA CCC; K624R, CCA ATG ATT GCA CCC CGG TTT GAA CCC; K624C, CCA ATG ATT GCA CCA CAA TTC GAA CCC). For cloning of the mutated DNA fragment the restriction enzymes XhoI and ClaI were used. The resulting cDNAs were cloned into the mammalian expression vector pCEP4 at SmaI and SalI sites.
The SP-GFP-APP 695 clone was prepared from plasmids pEGFP-C1 (CLONTECH) and pSP65-APP 695 (30). The GFP plasmid was restricted with PstI, blunt-ended with Klenow and religated to pEGFP-C0. The signal peptide of APP was amplified by polymerase chain reaction using pSP65-APP 695 as template and pairs of primers containing an engineered BamHI site (antisense primer: GGG GGC TAG CTC TAG ACA CTCGCA CAG CAG CGC ACT CG; sense primer: GGG GGA TCC ACC GGT ACC TCC AGC GCC CGA GCC). The DNA fragment encoding the signal sequence was amplified in pBluescript (inserted into BamHI/ SmaI sites) to pBS-SP. The GFP containing pBluescript plasmid pBS-SP-GFP-APP was cloned in three-way ligations using three inserts and restriction sites XbaI/ClaI of pBS (insert 1, pBS-SP fragment XbaI/ AgeI; insert 2, pEGFP-C0 fragment AgeI/KpnI; insert 3, pSP65-APP 695 fragment KpnI/ClaI). The DNA sequence encoding the GFP-APP 695 fusion protein was ligated into pCEP4 cut with NotI/XhoI.
The mutant construct SP-GFP-APP 695 /K624C was cloned into the pCEP4 vector (Invitrogen/ITC Biotechnology, Heidelberg, Germany) using the sites NsiI and PmlI. All constructs were verified by sequencing.
Cell Culture and Transfections-The pCEP4 vectors SP-GFP-APP 695 , SP-GFP-APP 695 -K624C, and the N-terminal c-Myc-tagged APP 695 and APP 695 -K624C (30,31) or the expression vector alone were transiently transfected into COS-7 cells in a LipofectAMINE Plus mixture (Life Technologies, Inc.). Stably expressing cell lines have been obtained by transfecting the human neuroblastoma SH-SY5Y with the plasmids K624C, K624R, and K625A containing the entire coding region of APP. COS-7 cells were grown in Dulbecco-Vogt modified Eagle's medium supplemented with 10% calf serum under an atmosphere of 5% CO 2 at 37°C. Human neuroblastoma SY5Y cells were grown at 37°C in minimal essential medium, supplemented with 10% calf serum, 2 mM L-glutamine, minimal essential medium nonessential amino acid mix and containing F-12. For stable expression, SY5Y cells were transfected with receptor cDNAs individually or in combination, selected with hygromycin (Life Technologies, Inc.), and maintained in medium containing 0.3 g/ml hygromycin. The properties of at least three different clonal lines were examined so that the influence of clonal variation on the observed phenotype could be determined.
Metabolic Labeling-Transfected cells of a 100-mm dish grown to 70 -90% confluence were washed with phosphate-buffered saline and then treated with 3 ml of minimal essential medium lacking methionine (Sigma) for 30 -45 min. Labeling was performed in 3 ml of minimal essential medium lacking methionine, 5% dialyzed HI-FCS (1-kDa cut off), supplemented with 300 Ci of [ 35 S]methionine (Amersham Pharmacia Biotech) and 5% CO 2 for 4 h. Alternatively, cells were incubated in medium lacking methionine and sulfate, supplemented with 5% dialyzed HI-FCS (1-kDa cut off) and 750 Ci of carrier-free [ 35 S]sulfate for 4 h.
Recombinant APP-(18 -350)-The construct, consisting of residues 18 -350 of APP from APP770, was expressed in Pichia pastoris as described (32). Briefly, culture supernatant containing recombinant APP from secreted expression was adjusted to 150 mM NaCl and applied onto a column of Q-Sepharose (XK-50/20; Amersham Pharmacia Biotech) equilibrated with buffer A (50 mM Tris-HCl, pH 6.8, 150 mM NaCl), eluted with 70% buffer B (50 mM Tris-HCl, pH 6.8, 1 M NaCl) at a flow rate of 10 ml/min and rechromatographed onto a second column of Q-Sepharose (XK 10/5). APP-containing fractions were pooled, adjusted to 250 mM NaCl, and loaded onto a Superdex 200pg (XK 16/60) column. Fractions were eluted at 0.5 ml/min in 1ϫ phosphate-buffered saline. The fractions eluting from an ion exchange and a subsequent size exclusion column were analyzed for APP content by Western blotting using monoclonal 22C11 as antibody. Protein concentrations were determined using the Bradford assay (Bio-Rad).
Cells were grown to confluence in 100-mm dishes, washed in phosphate-buffered saline, harvested, resuspended in lysis buffer (250 mM sucrose, 5 mM HEPES/NaOH, pH 7.4, 1 mM MgCl 2 , 10 mM KCl), and homogenized. Crude membranes were obtained by centrifugation at 800 ϫ g and resuspended in cross-linking buffer (150 mM NaCl, 20 mM HEPES/NaOH, pH 7.5). Freshly prepared cross-linker solution was added, and the suspension was incubated for 1 h at room temperature with continuous end-over-end rotation. To quench residual cross-linker activity, the solution was adjusted to 50 mM Tris-HCl (pH 7.5), and the incubation continued for 30 min at room temperature. To analyze the tetrameric, dimeric, and monomeric forms of APP, detergents Nonidet P-40 and Triton X-100 were added (to final concentrations of 2%) plus protease inhibitors (Complete tablets; Roche Molecular Biochemicals). The polyclonal antiserum 22734 was used to immunoprecipitate APP as described (33). Cross-linking of recombinant APP-(18 -350) was performed by using the nonreversible, homobifunctional cross-linking agent BS 3 or the cleavable cross-linker DTSSP (10 mM in water) in concentrations of 0.1-1 mM, depending on experimental requirements. To 10 g (200 g/ml in phosphate-buffered saline) of purified APP-(18 -350) freshly prepared cross-linker was added, and cross-linking was allowed to proceed for 45 min at room temperature. The reaction was terminated by the addition of Tris-HCl (100 mM, pH 7.5). Treated and control samples were analyzed by SDS-PAGE and immunoblotting with polyclonal APP antiserum or monoclonal 22C11 (34).
Antisera and Electrophoresis-Monoclonal APP antibody 22C11, which recognizes the N terminus of APP, and antibody W0 -2, which recognizes the amino-terminal region of A␤, have been described previously (34,35). Polyclonal APP antibodies 22734 were raised against recombinant APP from bacteria (22734) (33), or raised against yeast APP-(18 -350) (polyclonal 40090) as described earlier for recombinant APP from bacteria (20,36). Immunoprecipitation with polyclonal rabbit APP antiserum and monoclonal A␤ antiserum and immunodetection of APP were performed as described previously (33,37). For Western blotting of APP, samples were separated on 7.5% Tris-glycine or 7% Tris-Tricine gels and immunodetected with mAb 22C11. A␤ was separated by 4 -12% Bis-Tris gels (NuPAGE; Novex), and bands were quantified with the Fuji-Bas PhosphorImager system. Data are averages for at least three separate trials.
After electrophoresis, dried gels were autoradiographed to help the excision of radiolabeled APP from the gel. Protein was eluted overnight from gel slices in 0.2 M NH 4 HCO 3 with 0.1% SDS at 37°C. To increase the yield, gel slices were transferred into 0.2 M NH 4 HCO 3 , 30% acetonitrile with 0.1% SDS and repeatedly extracted for 3 h at 37°C. The pooled samples were dried under vacuum and stored at Ϫ20°C before electrophoresis.

RESULTS
In initial studies, we identified a conserved motif in the carbohydrate domain of recombinant APP for collagen binding and APP-APP interactions that were found to be modulated by N-terminal Zn(II), Cu(II), and heparin binding sites (8). To determine whether or not a monomer-dimer transition of APP also occurs in vivo, we analyzed APP dimerization and showed for the first time that APP exists as dimer and higher oligomers held together by multiple noncovalent protein-protein interactions.
Normally, cellular APP migrates as a single band with a molecular mass of 110-kDa upon gel electrophoresis, irrespective of whether the protein was loaded with or without reducing agent. To confirm that APP can oligomerize in vivo, APP 695transfected SH-SY5Y cells were labeled either with [ 35 S]methionine or [ 35 S]sulfate. Radiolabeled APP was immunoprecipitated from Nonidet P-40 and Triton X-100 extracts and resolved by nonreducing/reducing SDS-PAGE ( Fig. 1, A and B).
The addition of the cleavable cross-linker DTSSP to partially enriched APP resulted in strongly enhanced bands having the apparent molecular weight of APP 695 dimers and tetramers (Fig. 1A). Without the addition of cross-linker, such bands were visible as preexisting aggregates rather than functional dimers of APP obtained from crude membranes (Fig. 1A). This point of view is supported by the observation that cross-linked material did not all return to monomeric states after the cross-linker was reduced (Fig. 1B). Under nonreducing conditions, APP migrated at the monomeric molecular mass (110 kDa; Fig. 1A), whereas cross-linked APP migrated as dimers (ϳ190 kDa; Fig.  1A) and tetramers (ϳ400 kDa; Fig. 1A). To show that the APP dimers and tetramers were not covalently cross-linked and the cross-linking was not enzymatically induced by the tissue culture cells used, the different molecular forms of APP were isolated from the gel by preparative electroelution in a second experiment. APP complexes eluted from the gel were treated with DTT to cleave DTSSP bonds before running on a second SDS-PAG, and Fig. 1B shows APP forms of each eluted fraction from experiment 1 (Fig. 1A). As a control, size-matched gel pieces were eluted from untreated samples and failed to show any protein in electroeluted fractions from experiment 1 (data not shown).
In the presence of DTT and independent from the labeling technique, APP complexes were almost completely disrupted into monomers, with remaining original material (possibly resulting from preaggregated material) migrating either at the position of dimers or tetramers, respectively (Fig. 1B). Since we were unable to detect the presence of any proteins other than APP itself in the oligomers (even nonspecific cross-linking to other membrane proteins can be ruled out at 1 mM DTSSP), this suggests that oligomeric APP normally exists in SH-SY5Y cells as well defined dimers and tetramers. Most interestingly and as implied by the [ 35 S]sulfate labeling of APP complexes (Fig. 1, A and B), the oligomerization of APP occurs as early as in the ER, before [ 35 S]methionine-labeled APP complexes are exported and before maturation of N-linked oligosaccharides (38) and tyrosine sulfation of APP (20) occurs. It should be emphasized that the APP oligomers formed in the ER represent immature forms. This is most evident from To investigate the possibility that APP has oligomerization domains in addition to the collagen binding site (collagen binding peptide) of APP (8), we tested the oligomerization potential of APP-(18 -350) expressed in P. pastoris by dynamic light scattering and size exclusion chromatography (32). Using dynamic light scattering, we mainly observed dimers. The average diameter of aggregates was 4.6 Ϯ 0.1 nm at 20°C (data not shown). For verification of existing APP-(18 -350) dimers, size exclusion chromatography was performed on a calibrated Superdex 200pg XK16/60 fast protein liquid chromatography (FPLC) column, and fractions were analyzed for protein content ( Fig. 2A). The advantage of this technique is that it reveals the molecular weight of native proteins. The column was equilibrated with protein standards as indicated in Fig. 2C. Recombinant yeast APP behaved anomalously upon gel filtration. As  shown in Fig. 2A, the majority of APP eluted from gel filtration columns with a size of 113 kDa (fractions 33-36), which is much larger than expected for a monomer. A minor fraction of APP eluted at 267 kDa corresponding to tetramers (fractions 29 and 30). APP monomers were still observed with a molecular mass of 42 kDa (Fig. 2, A and C; fractions 41 and 42). This would be expected for a monomer-dimer equilibrium, which always has some unbound monomers. Because faster moving dimers dissociating from the monomer pool are not observed (as would be expected for a spontaneous monomer-dimer equilibrium), the equilibrium seems to be overbalanced in solution in favor of APP-(18 -350) dimers (and significantly less strongly in favor of APP-(18 -350) tetramers).
Analysis of the eluted material by SDS-PAGE (Fig. 2B) and Western blotting (data not shown) showed that the major peak contained APP-(18 -350) with an apparent molecular mass of 50 kDa. SDS-PAGE under reducing and nonreducing conditions indicated the absence of any intermolecular disulfidelinked homo-or hetero-oligomers (data not shown). This is in agreement with biochemical 2 and structural data (32), suggesting that all cysteines of APP-(18 -350) are involved in disulfide bridging. Taken together with the results from the SDS-PAGE (Fig. 2B), data from gel filtration experiments are consistent with the 113-kDa band being an APP-(18 -350) homodimer rather than a hetero-oligomer with another unknown protein.
To obtain additional evidence for the dimerization reaction of recombinant yeast APP-(18 -350), the protein was cross-linked by adding the homobifunctional cross-linker DTSSP or BS 3 , which both cross-link primary amines (i.e. predominantly lysine (Pierce)). No high molecular weight bands were observed without the cross-linkers (Fig. 3, lanes C). Additional bands at the molecular weight of APP-(18 -350) dimers and tetramers became visible when the samples had been incubated with the cross-linkers (Fig. 3). When 0.1 mM BS 3 or 0.1 mM DTSSP was added, two additional bands appeared at 100 and 200 kDa, representing dimeric APP-(18 -350) and tetrameric APP-(18 -350) (Fig. 3A). The distribution and intensity of those bands only slightly varied with the effective concentrations of BS 3 and DTSSP, implying that a stable monomer-dimer equilibrium was already reached at cross-linker concentrations of 0.1 mM (Fig. 3, A and B). Even in the presence of excess cross-linkers, the APP-(18 -350) monomer concentration remained constant, and a change in the oligomeric state of APP-(18 -350) due to a shift in the ratio of dimeric to oligomeric forms could not be noticed (Fig. 3, A and B). This indicates that collisional crosslinks can be ruled out and only cross-linking of stable dimers was observed.
Thus, we conclude that at least two sites are involved in APP dimerization. The ectodomain of APP 695 showed an intrinsic ability to homodimerize through the collagen binding site in our earlier study (8), and APP-(18 -350) (containing all 18 cysteines of full-length APP) has an even stronger dimerization potential than APP from SY5Y cell membranes (see band intensities in Figs. 1 and 3) suggesting a zipper-like mechanism for dimerization of full-length APP.
To corroborate the oligomeric state of cellular full-length APP and to analyze the influence of oligomerization on intracellular APP processing (i.e. amyloid A␤ production), we wanted to generate covalently stabilized APP dimers with an intermolecular disulfide bond. We introduced a single unpaired Cys in the extracellular domain in the region immediately adjacent to the transmembrane domain of APP 695 (residues 625-648, encoded by exon 17). Because transmembrane domains are thought to emerge from the membrane in an ␣-helical conformation, the projected APP 695 -K624C substitution (i.e. A␤ residue 28) was analyzed to fit with the predicted amphipathic interface in the APP juxtamembrane domain. The wild-type and the Cys-mutant APP 695 , which were overexpressed in SY5Y cells by stable transfection, all migrated at a relative molecular mass of about 110 kDa in SDS-PAGE under reducing conditions (Fig. 4A). In contrast, all of the Cys-mutant proteins, but not the wild-type protein, migrated predominantly at about 220 kDa under nonreducing conditions (Fig.  4A, lane APP 695 -K624C, ϪDTT). All mutant constructs investigated, K624C, K624R, and K624A, yielded the same amounts of secreted APP as in the wild-type transfected cells (data not shown), but dimers were only formed from the K624C construct. This indicates that the APP constructs were correctly expressed, fully glycosylated, and transported and that the Cys mutant efficiently dimerized in vivo by intermolecular disulfide formation (Fig. 5). Thus, APP 695 containing the K624C mutation in the extracellular domain dimerized constitutively via a disulfide bond, provided that the K624C mutant was most effective in stabilizing the APP dimer (Figs. 4A and 5). When corrected for the amount of secreted APP precipitated from the same sample, monomeric A␤ was produced to a similar extent from all APP forms (Fig. 4B, lanes APP 695 -K624C and APP 695wt, ϪDTT). APP and A␤ levels from three independently generated APP 695 and APP 695 -K624C cell lines were densitometrically calculated in the presence and in the absence of DTT, and the ratio of A␤/APP was determined. The ratio of A␤/APP obtained from APP 695 cells was used as control and set at 100%. The evaluation resulted in an increase of monomeric A␤ under reducing conditions of 712.8 Ϯ 10, 674.7 Ϯ 12, and 692.8 Ϯ 14%, yielding a mean value of 693.4 Ϯ 12%. Thus, after conversion of dimeric A␤ to monomeric A␤ under reducing conditions (Fig. 4B, lanes APP 695 -K624C and APP-wt, ϩDTT) the A␤ level was found to be increased by a factor of 7. By pulse-chase labeling experiments, we found that the expression level of APP 695 and the K624C mutant did not vary; nor was the half-life time of either of them impaired (data not shown).
How could the 7-fold increased level of A␤ then interfere with APP processing? This interference could either be due to the APP 695 -K624C mutant construct facilitating ␤-cleavage and attenuating ␣-cleavage, an overall increased expression of APP 695 -K624C, or inhibited A␤ clearance. To investigate if the ratio of ␤-secreted versus ␣-secreted soluble APP 695 derived from the K624C mutant was changed, Myc-tagged APP 695 -2 L. Hesse, manuscript in preparation. K624C was immunoprecipitated with rabbit polyclonal anti-c-Myc antibody (33) and analyzed by monoclonal antibodies 22C11 (recognizing equally well sAPP␣ and sAPP␤ due its epitope at the N terminus of APP (34)) and W0 -2 (recognizing exclusively sAPP␣). Since we were unable to detect differences between the total amounts of APP secreted (22C11 immunoreactivity of APP 695 versus APP 695 -K624C mutant) and between sAPP␣ and sAPP␤ immunoreactivity (data not shown), we favor the possibility that dimerization of A␤ inhibits intracellular clearance. Consequently, it is attractive to postulate that APP dimerization facilitates the stabilization of A␤ dimers that could very well represent a preliminary stage of beginning fibrillogenesis. A␤ dimers have already been shown to exist under physiological conditions (39).
The precipitation of soluble A␤ dimers from cell culture supernatant indicates that the disulfide bonds were maintained throughout the transport to the cell surface (Fig. 4B). The high yield and the release of A␤ dimers into the cell culture supernatant indicates that in the course of folding, disulfide bridges rearranged spontaneously by intramolecular thiol-disulfide exchange in the correct way (40).
We confirmed that the APP oligomers found were actually homodimers, because Myc-tagged APP 695 -K624C could form homodimers (Fig. 4C; lane myc-K624C) that were never observed with GFP-tagged APP 695 -K624C (Fig. 4C; lane GFP-K624C), possibly because of steric hindrance in the N-terminal dimerization domain of GFP-APP 695 -K624C. The only heterodimer observed was formed between Myc-tagged APP and GFP-tagged APP 695 -K624C (Fig. 4C; lane myc-/GFP-K624C). Immunoblotting analysis showed that the Myc-tagged APP 695 -K624C was expressed at a higher level than the GFP-APP 695 -K624C molecule but that the GFP-APP/Myc-APP heterodimer was preferentially formed in doubly transfected cells (Fig. 4C; lane myc-/GFP-K624C). Thus, our results also indicate that a dimer may only form when monomers in the proposed dimer are oriented in a particular geometry with respect to one another. The extended N terminus of GFP-APP 695 -K624C is inhibitory for dimerization with another GFP-APP 695 -K624C molecule but not with another Myc-APP 695 -K624C (Fig. 4C), and although the predicted size of GFP-APP 695 -K624C is higher, it migrates faster than wild-type APP. Both observations can be explained by insights into the conformational changes of the protein backbone of APP that we gained when we analyzed expressed APP fusion proteins in Escherichia coli (41). We found the characteristic aberrant migration of different naturally occurring isoforms of APP mainly to depend on aspartic and glutamic acid residues within the N-terminal domain of APP rather than on post-translational modifications or the length of the protein (i.e. the number of amino acids).

DISCUSSION
By three lines of evidence (cross-linking experiments, size exclusion chromatography, and mutational analysis), we could demonstrate that, under native conditions, a certain part of native APP exists in a monomer-dimer or monomer-tetramer equilibrium. For the first time, we present evidence that cellular APP is capable of forming noncovalent homodimers and tetramers. By chemical cross-linking, we show that homodimerization of full-length APP is extensively enhanced in membranes of SY5Y cells (Fig. 1). Most interestingly, the analysis of post-translationally modified APP from SY5Y cells reveals that APP dimers appear to assemble in the ER shortly after the synthesis, suggesting that the homodimeric state could even be an essential prerequisite for its sorting, in the trans-Golgi-network and to secretory vesicles.
Most likely, the N-terminal domain is both necessary and sufficient for homodimerization, since the assembly of oli-  (18 -350) to be higher than that reported for PrP (i.e. 3.9 ϫ 10 8 M Ϫ1 at 37°C) that also forms dimers but only as a fully post-translationally modified protein (42).
Two highly conserved regions of APP, the N-terminal APP residues 91-111 3 and the C-terminal APP residues 448 -465 representing the collagen binding site (8), are of critical importance for the regulation of homo-oligomerization of full-length APP. 3 Thus, dimerization may occur at physiological pH values in the lumen of the secretory pathway from the ER to the secretory vesicles in a zipper-like mechanism. This can be concluded from the observation that the N-terminally extended and high molecular weight form of GFP-APP 695 -K624C affects the dimerization. For conformational differences between the GFP-APP 695 -K624C and the Myc-APP 695 -K624C, the GFP-APP fusion part was inhibitory for another GFP-APP to bind but not for the shorter Myc-APP. Thus, we conclude that in the dimer formation, the interaction mediated between the N-and C-terminal domains of APP behave in a cooperative manner, with the N-terminal one being a prerequisite for efficient interaction and the C-terminal domain linking two preexisting N-terminally bound APP dimers to one another.
In accordance with this model, the Cys-mutant K624C of APP 695 formed disulfide bridges, which rearranged spontaneously, possibly in an extending formation of the proposed linear zipper-like array. The specific ability of a disulfide cross-linking to form at the juxtamembrane domains of APP implies that the juxtamembrane domains of APP molecules are closely apposed in the receptor dimer. Thus, we suggest that dimerization of APP through at least two separate effector sites, the N-terminal domain and the collagen binding site (8), induces an extended third contact between juxtamembrane and intramembrane ␣-helices enabling a spatial proximity of the A␤ regions of two APP molecules.
The disulfide bond linking of two K624C APP 695 molecules could only be generated through oxidation in the ER. The ER lumen is unique among the various compartments in the eukaryotic cells because it provides an oxidizing environment for the disulfide bond formation with the help of the disulfide isomerase (43). The disulfide bond formation occurs co-translationally together with folding and oligomerization of proteins within the ER and does not interfere with the secretory pathway. APP dimers must also have acquired the correct folding pathway, since the dimerized APP 695 -K624C mutants were transported from the ER to the Golgi, glycosylated, sulfated, and directed into the secretory pathway. As expected for a successful processing of dimerized APP 695 -K624C within the ␣-secretory pathway, only post-translationally modified monomeric forms of APP 695 were observed in the cell culture supernatant from APP 695 -K624C-transfected cells. This is due to the fact that secreted APP is derived from dimerized mutant APP 695 -K624C, which is monomerized by the ␣-secretory cleavage that occurs C-terminally to APP residue 612 (APP 695 numbering) but N-terminally to the site-specific cross-link (APP residue 624) (Fig. 5). The arrest of transport of N-glycosylated mutant K624C APP 695 mediated by a temperature block at 15°C (data not shown) and the failure of other Cys mutants to form disulfide-linked APP dimers (data not shown) indicate that covalent dimerization of K624C is not simply due to ER retention of misfolded protein.
The disulfide bond within the amyloid domain of APP 695 -K624C, 12 residues C-terminal to the ␣-secretase site, allowed us to quantify amyloid released as dimers from APP 695 -K624C. The increased level of soluble A␤ was made visible after reduction of A␤ dimers breaking the cystine at APP residue 624 (i.e. A␤ residue 28) by adding DTT and measuring the increase of monomerized A␤. The APP 695 -K624C mutant analyzed in SY5Y cells produced a 7-fold increase of A␤. This is in agreement with the presently valid maximum of a 6 -8-fold increase of A␤ released by cultured cells that were transfected with APP bearing the so-called Swedish mutation (Lys to Asn at residue 595 plus Met to Leu at position 596) (44). According to an earlier hypothesis that covalent cross-linking of A␤ dimer is the first step of amyloid formation (45) and SDS-stable A␤ dimers occur in vivo (46), initially formed A␤ dimers from APP dimers may further aggregate into amyloid fibrils also in vivo.
A dimerization-mediated regulation of APP metabolism is most likely to be physiologically relevant for the following reasons. Although our current model implies that APP dimerization does not serve as a signal to divert APP into the ␤-secretory pathway, APP may exist in the cell as populations of monomers and dimers in equilibrium. Ligand binding, for instance copper or zinc (8), can direct cellular APP into the ␣-secretory pathway (33,47). According to our earlier studies and to the data presented here, we are currently investigating the inhibition of dimerization as a possible mechanism. Proteins interacting with the intracellular C-terminal domain of APP, such as the neuronal adaptor protein X11␣ (48) or presenilins, also could modulate the extent of dimerization of membrane-bound APP. Upon internalization of APP monomers, such forms of APP would be less prone to A␤ production than internalized dimers.
Dimerization of APP matches a proposed signaling function of APP. APP is part of a G 0 protein-centered complex that transduces extracellular signals to the cytoplasm and the nucleus (49). If one assumes that binding of antibody 22C11 to the extracellular domain of APP (residues 66 -81) (34) exerts its 3 G. Multhaup, manuscript in preparation. The transmembrane domain is indicated in yellow, and A␤ is shown in red. The part of A␤ that is inserted in the membrane is enlarged. Amino acid residues of A␤ are given in one-letter code. Note that the K624C mutation is located at the juxtamembrane position. activation of G 0 by a ligand mimetic mechanism (8,50), one must argue that monomeric APP may function as a G proteincoupled receptor. This also indicates that as long as APP is not needed in its functional monomeric form, APP may exist as dimers in the absence of ligands. Thus, APP appears to be negatively regulated by dimerization as has been described for receptor-like protein-tyrosine phosphatases (51).
One of the questions that remain is whether other APLPs dimerize like APP. The conservation of the cysteine-rich Nterminal domain is consistent with the possibility that all APLPs can dimerize and the variable residues in this region could provide the necessary dimerization specificity. To identify novel risk factors for Alzheimer's disease and possibly inhibitory compounds, it will obviously be important to determine whether APP and APLPs can heterodimerize and to determine how APP dimerization and increased A␤ production are linked with the signaling function of APP.